The project aims to design an interactive queries function of SQL by connecting the database from
Azure to evaluate the KPI performance.
Local Database (DB) connection to Azure is executed by using Azure Database for MySQL single
server(use Migration Wizard). In codespace, we can access Database which is exactly the same as
the Azure database, and with this, I created a Datawarehouse and loaded all the needed data from
the DB. Interactive queries can be perfromed in the codespace with the SQL interface. The goal
of this project is to let KPI evaluation be more efficient and updatable by connecting to a
cloud Database.
The goal of this project is to develop a tool that can distinguish between fake and authentic
images using a pre-trained MobileNet model. With advancements in AI-generated images, the
ability to accurately detect fake images has become increasingly important. The project aims to
provide a useful and reliable solution for identifying fake images and helping people verify the
authenticity of visual content.
Generative models have been widely applied to the realm of natural lanauage. Automatic generation
or Images or texts has been prevalent, while music generation is a relatively niche area that we
aim to delve deeper into via this project. We explore both the conventional ~Transformer and
Generative Adversarial Network (GAN)-based models to evaluate their quality and
performance in the context of conditional music generation.
The web application is built to return JSON payload containing the news that are relevant to
what users want to read. Overall, it is developed using AWS cloud 9 as the environment and
queries from the Atlantic news website. The microservice was containerized in AWS Elastic
Container Registry (ECR) and pushed to production using AppRunner and FastAPI. Continuous
Integration is enacted through Github Actions and Continuous Deployment is performed through
configuring Build Server to deploy changes on build using AWS ECR and Code Build.
The project aims to build a serverless web application that is able to reuturn a sentiment report
without a server, and perform Continuous Integration through Github Actions and configure Build
Server to deploy changes on build (Continuous Delivery) using AWS Lambda, API Gateway and S3
bucket.
The project aims to design an interactive Line Bot that can help track the to-do list. Line is a
message platform where people in Taiwan, Japan and Korea use in their daily life. It deployed
AWS Lambda Function that was built to connect dynamoDB (using boto3) and Line (using the
line-bot-sdk). To establish the connection between users and channels,the Lambda Function was
also connected with AWS Gateway using the REST API, which would allow the messages sent from
Line users to store in dynamoDB, and allow the information to be pulled out from dynamoDB.
This HackDuke project aims to build an innovative and intuitive web app for sustainability
fashion, which achieved a third-place in the environmental track. An image-to-text neural
network model, natural language processing were adopted to help users identify the estimate
hidden electricity, water, and CO2 costs of a piece of clothing by simply scanning the clothing
label. The web app allows users to conceptualize the environmental impact of their clothing
purchases, and stay informed when making purchase decisions. The app was deployed on Heroku and
is currently paused due to high costs of running it.
Using 400M+ rows of data, the project examines the effectiveness of interventions in the State’s
Opioids policies in lowering the volume of opioids prescription and the rate of drug overdose
mortality. Pre-Post analysis is first conducted to examine the difference in opioids
prescription and mortality rate between pre-policy and post-policy era. However, even if the
trend is different, we cannot be sure the difference is caused by the policy and not other
factors. Thus, differnece-in-difference analysis is conducted. This method allows us to compare
the change in opioids prescription and mortality rate in the state where the policy was enacted,
with a hypothetical parallel universe where the policy was not enacted. Overall, the project
provides evidence-based recommendations on the effectiveness of the State's Opioids policies,
and to inform future policy decisions on opioid use and abuse.
The project aims to evaluate whether company descriptions on LinkedIn are
sufficient to identify the type of business. The current study focuses on three predominant
industries: Financial Services, Hospitals & Health Care, and IT Services & IT Consulting, while
it can be scaled to incorporate additional business sectors for broader use
cases. The study deploys BERT as a discrimitive model and Multinomial Naive Bayes model as a
generative model to compare their effectiveness of classifying company's industry based on the
company
descriptions on LinkedIn. Synthetic data was also generated based on Naive Bayes model in order
to assess how the two models perform under real conditions and under the conditions for which
the generative method was designed. Results suggest Naive Bayes have a slightly better
performance in classifying the company industry than BERT model, even though both achieved high
accuracy.
The project built a c command line tool in Rust that allows users to quickly record their diary
with a time stamp. Rust's memory safety features and static typing could ensure that the tool is
robust and free from common programming errors such as buffer overflows or null pointer
dereferencing. Rust's concurrency features could also be leveraged to allow multiple users to
use the tool simultaneously without data corruption or synchronization issues. The resulting
command line tool would be efficient, fast, and reliable, making it a useful tool for users to
easily record and recall their diary entries.